Schema types in 2026 — semantic validity versus search features

Edition: 2026.09.09 · Status: researched replacement of the legacy entry point
Primary sources: S01 S20 S21 S22 S23 in the source register.
Preservation: The complete prior document remains byte-for-byte in 99-Originals/Framework/framework-schema-types-2026.md. This replacement deliberately removes unsupported universal rules; it does not claim to have tested a client website.

A type catalog is not a benefit catalog

Schema.org vocabulary and search-engine feature support are separate layers. A valid type can describe content even when Google does not offer a corresponding rich result. Feature eligibility additionally depends on required properties, policy, content and availability. Appearance is not guaranteed. [S20]

Current planning matrix

Family	Use in this library	Acceptance boundary
Organization / WebSite	Model genuine identity and site relationships	Real identifiers and visible business facts; no knowledge-panel promise.
Article / Person	Model actual editorial content and authorship	Real author/content relationship; no fabricated credentials.
BreadcrumbList	Describe the real navigational hierarchy	Match meaningful navigation and valid URLs.
Product / Offer	Describe a genuine product and commercial offer	Follow the current feature guide and reconcile page, feed and checkout.
LocalBusiness	Describe a legitimate applicable business entity	No invented storefronts, addresses or local staff.
Review / AggregateRating	Describe authentic qualifying reviews	Apply current feature policy; no placeholder ratings.
FAQPage	Optional vocabulary where justified	Google FAQ rich results retired May 7, 2026.
HowTo	Optional vocabulary for genuine instructions	Google HowTo rich results retired September 2023.
Other types	Evaluate individually	Consult the current specific consumer guide before promising a feature.

FAQ and HowTo lifecycle facts are from S01 and S21. Do not copy an old rich-result checklist into a 2026 proposal. Do not assume that removal of one type implies every neighboring feature, such as all book-related integrations, is also retired.

Model only supported facts

Use the most accurate applicable type without inventing a subtype to appear sophisticated. Populate properties from controlled records. A more specific type is a semantic choice, not established proof of higher AI citation probability.

For commerce, optional ratings are not mandatory filler. Keep prices, currency and availability current. Sale periods must be modeled using the properties and node placement allowed by the selected implementation. [S23]

For reviews, apply genuine-experience and disclosure requirements and the self-serving LocalBusiness/Organization restriction. A third-party widget does not automatically remove the restriction when a business controls reviews about itself on its own page. [S22]

Generation and testing

Generate JSON using a serializer rather than string concatenation. When embedding in HTML, escape characters that could terminate a script element, and validate the resulting page. Never expose secrets or private catalog fields in a public graph.

Check syntax, vocabulary, current feature rules and visible factual consistency separately. Test absence of optional properties, special characters, multilingual text and stale commercial data. Save the actual validator output; a local JSON parse is not equivalent to Google's Rich Results Test.

The example graphs in Schema.org contain fictional data and do not claim feature approval. Use schema lifecycle for maintenance and commerce consistency for product pipelines. Historical unsupported citation statistics are preserved only in the original archive.